In a landmark move that fundamentally alters the landscape of AI hardware, Nvidia has officially announced a $2 billion strategic investment in Marvell Technology. The deal, unveiled on March 31, 2026, aims to integrate Marvell’s custom silicon and advanced networking expertise directly into Nvidia’s “AI Factory” ecosystem.
This partnership marks a rare shift for Nvidia, moving from its historically proprietary strategy toward a more collaborative, heterogeneous computing architecture to maintain its dominance.
1. The Core Strategy: “NVLink Fusion”
The investment centers on Nvidia NVLink Fusion, a new rack-scale platform designed to allow customers to build semi-custom AI infrastructure while remaining fully compatible with Nvidia’s software and hardware stack.
- Custom XPUs: Marvell will provide specialized accelerators (XPUs) and scale-up networking gear that can plug directly into Nvidia’s systems.
- The Supporting Stack: Nvidia will supply the “glue” for these custom builds, including its Vera CPU, ConnectX NICs, BlueField DPUs, and Spectrum-X switches.
- Flexibility for Giants: This move is largely seen as a way to keep “hyperscalers” (like Amazon and Meta) within the Nvidia ecosystem even as they develop their own internal AI chips.
2. Solving the “Bandwidth Bottleneck”
A major portion of the $2 billion is earmarked for the joint development of Silicon Photonics and Optical Interconnects.
| Technology | Objective |
| Silicon Photonics | Replacing traditional copper wiring with light-based data transmission to slash energy consumption. |
| 1.6T Optical DSPs | Doubling the bandwidth between server racks to handle the massive data requirements of GPT-5+ class models. |
| AI-RAN Ecosystem | Collaborating on Aerial AI-RAN to turn 5G/6G telecommunications towers into distributed AI edge data centers. |
3. Market Reaction: Marvell Stock Soars
The announcement sent shockwaves through the semiconductor market, with Marvell emerging as the biggest winner of the day.
- Stock Performance: Marvell (MRVL) shares surged 13% on Tuesday, with analysts raising price targets to $135 per share.
- Nvidia Status: Nvidia’s stock also saw a modest lift, as investors cheered the company’s proactive approach to neutralizing the threat of “custom silicon” rivals.
- Revenue Projections: Marvell now expects its AI-related revenue to approach $15 billion by fiscal 2028, driven largely by this deepened Nvidia relationship.
4. Why Nvidia Chose Marvell
While Nvidia has flirted with other ASIC (Application-Specific Integrated Circuit) partners, Marvell was chosen for three specific reasons:
- Optical Leadership: Marvell is a global leader in high-speed optical DSPs, which are critical for scaling AI clusters beyond 100,000 GPUs.
- Proven Track Record: Marvell already works with AWS (on Trainium/Inferentia chips), proving it can deliver high-performance custom silicon at scale.
- Defensive Moat: By investing $2 billion, Nvidia effectively “gates” one of the best custom-chip designers, making it harder for competitors like Broadcom to form similar exclusive alliances.
5. The “Inference Inflection”
In a joint CNBC interview, Nvidia CEO Jensen Huang and Marvell CEO Matt Murphy noted that the world is moving from “training” AI to massive-scale “inference.”
“The inference inflection has arrived,” Huang stated. “Together with Marvell, we are enabling customers to build specialized AI compute factories that can generate tokens at a scale previously thought impossible.”
